4. LITERATURE OVERVIEW
4.3 Oxygen reduction on non-precious metal catalysts
Niedrach of General Electric (GE) for the Gemini missions [28]. Since the early days, the large amount of Pt needed to produce the GE fuel cells was noted to be a significant problem towards commercialisation [28]. Recently, PEFCs have seen a lot of success: major automotive companies such as Toyota, Hyundai, Honda, Ford, Chevrolet and Mercedes-Benz have all come out with their own PEM fuel cell vehicles and Plug Power, a company producing hydrogen-powered forklifts, posted more than $230M of revenue in 2019 [29]. However, all of these commercial solutions are still powered by Pt-based catalysts. So, what is wrong with using platinum? The global platinum supply is estimated at 69,000 tons and 192.5 tons were produced in 2019 [30,31]. A Toyota Mirai (the most produced fuel cell electric vehicle, FCEV) needs 35 grams of Pt for its fuel cell, meaning that if all the Pt currently produced was used for FCEVs, 5.5 million of such vehicles could be produced per year. The global car fleet was estimated to be over 1 billion in 2010 [32], so to replace the entire car fleet with Toyota Mirais (or vehicles using the same amount of Pt in the PEFC) would take over 18 years at the current pace. This requires that 35,000 tons (before taking into account the losses of making catalysts from the mined Pt) of the 69,000 tons available would need to be dedicated to FCEVs, which means that the recycling rate for Pt would need to be near 100% in the long term. Another thing to consider is that currently, the price for a Toyota Mirai is $58550 compared to $24525 for a Toyota Prius (an identical vehicle with a hybrid drivetrain). Clearly this is not a competitive
price. The main problem stems from the fact that an average Pt ore contains only 4 g of Pt per ton of ore and that most of the mining is done in deep mines in South Africa, driving up the price even more. Because the mining is mostly concen-trated in only one area, there are also concerns about the stability and elasticity of the supply should the demand increase notably [5]. Pt-based catalysts have seen a tremendous increase in both specific activity and mass activity (by mass of Pt) in the recent decade as well [33–35], however incorporating these advancements made on the catalyst level to full-size fuel cell stacks has proven a challenge [36].
Even the most optimistic forecasts set the ultimate Pt loading at 0.1 g kW–1 [34], which would mean that 1/7th of the whole Pt on Earth would need to be dedicated to PEFC technology (not taking into account recycling and catalyst synthesis losses).
Clearly, in the long term, Pt-based catalysts are not the solution for PEFCs in FCEVs and even the US Department of Energy has included moving on to NPMC materials in their PEMFC roadmaps in the last few years [6,7]. The search for Pt replacements for oxygen reduction also has a long history, even predating PEFCs.
As mentioned, the ORR also takes place in the respiratory system of mammals.
Due to the low electrical conductivity of biological systems, the reaction rate is much lower than what is needed for a fuel cell, however. In 1964, Jasinski pub-lished the first study using a bio-inspired cobalt phthalocyanine (CoPc) catalyst deposited on Ni (as a highly conductive substrate) [8]. The next year he improved on his results considerably by using a carbon black instead of Ni as the substrate [37]. The carbon black had the advantage of having a much larger surface area, allowing for more of the CoPc to be adsorbed and therefore, more electro-catalytically active sites to be exposed. Nowadays, this strategy of using a carbon support to increase the utilisation is key in both precious and non-precious metal catalysts [33,38,39]. This was improved upon by introducing a high-temperature treatment which improved the contact between the MN4 macrocycles and the substrate (but also changed the nature of the active site, which will be discussed further on)[40]. The next important enhancement was done in 1989 by Yeager and co-workers, who replaced the expensive MN4 macrocycles with a mixture of simple nitrogen source (polyacrylonitrile, PAN) and simple metal salts (Co or Fe acetate). These mixtures were pyrolysed at high temperatures (300–950 °C) to create some of the best NPMCs at that time [41]. However, unlike the physically adsorbed macrocycle-based catalysts, which had a well-defined structure, these new heat-treated materials proposed the important question about the exact nature of the active sites for ORR after pyrolysis, which has only begun to be answered in the recent years [26,42–45].
strength of the material. Traditionally, carbon materials with a high surface area such as Vulcan carbon XC72 and Black Pearls BP2000 have been widely used as a support material for heterogeneous catalysts, notably as carriers for Pt nano-particles [33,46]. However, carbon by itself is not very active towards the ORR.
To create active sites for the ORR without using platinum, doping it with transition metals and/or nitrogen has been the most successful strategy so far [33,47–49]. In the case of unpyrolysed transition metal phthalocyanines, the activity of metals decreases in the following order: Fe > Co > Ni > Cu > Mn [50], but most catalysts today are pyrolysed, therefore the nature of the active site is very different from a pure metal phthalocyanine [51,52] and the activity trends also different. Before going into discussion about the exact nature of the active site, it is important to understand how these materials are made.
The most common methods for creating M-N-C catalysts are:
1) The doping method, where a synthesis mixture, comprising of a carbon carrier, nitrogen dopant (either a polymer or smaller molecule with a high density of N atoms) and a metal source (commonly a cheap salt) is pyrolysed to dope the carbon material. The nitrogen and metal source can be combined (for example, a macrocycle containing the desired metal). The precursors are commonly mixed either in liquid (sonication, stirring in a solvent) [I–IX] or solid phase [VII–XIII] (grinding, mixing, milling) after which the mixture is pyrolysed at high temperatures to fuse them together, which changes the chemical nature of the dopants and the substrate, creating active sites for ORR [53–57]. The advantage of this method is the use of a pre-existing carbon, which defines the structure, porosity and degree of graphitisation of the final catalyst.
2) The in situ doping method, which is similar to method 1, but the doped carbon network is created during the pyrolysis from carbon, nitrogen and metal sources (these can either be the same source, such as a Fe-doped ZIF-8, for example, or from different sources such as a carbon-based polymer, nitrogen-containing molecule and an iron salt). This method can facilitate (in most cases) a higher concentration of nitrogen and metals compared to method 1, but it is more difficult to control the structure and porosity of the final catalyst.
This is commonly alleviated by using a precursor (or precursors) which has a well-defined structure, such as metal-organic frameworks (MOFs) [44,45,58–
61], metal macrocycles [62,63], macrocyclic aerogels [64] or polymers [65,66].
3) The hard-template method (also called the sacrificial support method, SSM), which incorporates the same strategy as 1) but in addition, a hard template (commonly made of silica) is used to confine the precursors during the pyrolysis. This allows to define the final structure of the catalyst by using a template which is stable during the pyrolysis (such as silica) [67–69]. There-fore this method combines two of the positive sides from methods 1 and 2, but has the disadvantage of having to remove the template after pyrolysis, a process which can also modify the resulting catalyst by etching some of the metal (not always a negative side as the final catalyst can therefore be more stable [67]).
In many cases, these methods are combined, i.e. the catalyst produced by the soft-template method is also doped afterwards by either adding more nitrogen-containing compounds and pyrolysing the mixture [70] or pyrolysis in ammonia [44,45,61]. Acid washing is also common addition to all of the methods to remove inactive metal species, which could contaminate the fuel cell during operation, but carries the disadvantage of having to undertake a secondary heat-treatment to deprotonate the N-groups on the surface (which can bind anions and therefore decrease the activity of the catalyst) [71].
The most important questions in any research done on catalytic reactions are:
1) What is the chemical nature and structure of the catalytic centre?
2) How is the reaction proceeding on that centre?
These two questions have been the centrepiece of research done on M-N-C electrocatalysts for ORR ever since pyrolysed macrocycles were used for ORR electrocatalysis in 1976 [40]. Two main types of active sites towards ORR have been identified in M-N-C catalysts: M-Nx single metal-atom sites, where the iron atom is coordinated to multiple (usually 4) nitrogen atoms [43,44,72–74], or metallic iron and/or iron carbide particles covered by nitrogen-doped graphitic carbon layers, labelled NC@M [25,45,75–79] (Figure 4). In addition to the metal-based catalytic centres, M-N-C catalysts also contain NxCy active sites, which will be further discussed in the next chapter. M-Nx sites are generally considered to reduce oxygen via a 4-electron pathway, while on NC@M sites the ORR is thought to proceed either via the 2+2 or 2×2-electron pathway, with the underlying metal stabilising the intermediates [43,45,61].
The exact nature of the M-Nx sites most active towards the ORR is extremely difficult to determine due to the plurality of even this one type of active site in a given catalyst. Due to the imprecise method of synthesis (high-temperature pyrolysis), all catalysts thus far have had multiple types of active sites present.
Commonly these active species are identified in a catalyst by X-ray photoelectron spectroscopy (XPS), Mössbauer spectroscopy or X-ray absorption spectroscopy (XAS). However, with all of these methods, the signal from different M-Nx sites (and in the case of XPS, also other nitrogen moieties) is overlapping and thus a large part of the identification is the deconvolution and fitting of peaks [27]. In perfect systems such as graphene direct atomic-level imaging has also been achieved [74], but most of the catalysts are an amorphous mess compared to graphene and determining how many and what atoms are bound to the metal centre is rather speculative with current imaging capabilities.
Due to the difficulties in synthesising a catalyst with a single type of active site, the exact reaction mechanism on this type of catalysts has also been difficult to determine. Figure 5 shows a general scheme of the ORR taking place on a M-N-C (in this case, Fe-M-N-C) catalyst with a multitude of active sites in acidic conditions [27].
Figure 5. ORR on a M-N-C catalyst with different active sites: S, S*, S1
and S2 [27].
As shown, Fe-Nx sites catalyse a direct 4-electron reduction of oxygen to water or the 2x2e– reduction, where both of the steps take place on either the same active site or another S* site. On nitrogen moieties without any coordinated iron, the ORR undertakes the 2+2e– reduction with the first steps of oxygen adsorption and reduction to H2O2 taking place on pyrrolic or graphitic nitrogen and the second step on pyridinic nitrogen (more discussion on this will follow in subsection 4.3.2) In alkaline conditions, as discussed, the reaction preferably takes place in the outer Helmholtz plane (OHP) and is thus tilted towards the 2-electron mechanisms [25,27,80]. However, it must be noted that the surrounding carbon also has an influence on the active sites and can thus change the selectivity.
Similar moieties have also been characterised in the case of Co and Mn [70,81].
Nevertheless, the main goal of the research into M-N-C catalysts has always been to replace Pt on the cathode of PEMFCs. Looking at a comparison of a state-of-the-art M-N-C catalyst and a state-state-of-the-art Pt/C catalyst [82] (Table 1, updated for 2019) reveals that while by electrode area the activity is similar, the mass activity is two orders of magnitude lower. This also means that the electrode mass, and more importantly, volume required to reach the same power as the Pt/C catalyst will be much larger. In a confined space such as an automobile, making the electrode 100 times thicker is obviously not possible. Thickening the electrode also creates considerable issues with O2 transport in the catalyst layer.
Table 1. Comparison of state-of-the-art Pt/C and M-N-C catalysts in PEMFCs at 0.9 V [39].
Catalyst Loading Mass activity Catalyst activity
Pt/C 0.1 mgPt cm–2 443 A/gPt 44.3 mA cm–2
(CM-PANI)-Fe-C 6.8 mg cm–2 5.2 A/g 36 mA cm–2
There are two main reasons the mass activity of M-N-C catalysts is so much lower: site density (SD) and turnover frequency (TOF). In a recent cross-laboratory study on some of the best Fe-N-C catalysts for the ORR, the catalyst with the highest site density had a SD of 0.6×1019 (accessible) sites cm–3 [83] compared to 3.2×1020 sites cm–3 for Pt/C [24]. Looking at TOF, the TOFs reached in the same study [83] were 0.5–7e– site–1 s–1, while commercial Pt/C is known to have a TOF value of ≥25e– site–1 s–1 at 0.8 V [24]. Another study reported a TOF of 0.17 e– site–1 s–1 for a Fe-N-C catalyst [84].
Reaching a site density comparable to Pt/C with an atomically dispersed
atoms tend to agglomerate into particles, creating NC@M sites, which are less active. Therefore, the recent strategies for increasing site density have focused on separating the metal atoms prior to and during the pyrolysis step, which is accom-plished either by dispersing them in a MOF, in complexes with large ligands or anchoring using SiO2 [60,64,67,85]. The second issue, TOF, is somewhat more difficult to improve without knowing the exact nature of the active sites. Still, a lot of progress has been made in the case of Fe-N-C catalysts by tuning the dz2
orbital electron density of the central Fe atom [27,86]. By increasing the defectiveness of the carbon material surrounding the Fe-N4 centre, it is possible to increase its electron withdrawing nature, which in turn lowers the electron density on the dz2 orbital so that it can be tuned to bind intermediates just strong enough. However, by introducing too many defects into the carbon material can make the Fe-N4 site bind oxygen too strongly and has another very important downside: it decreases the catalysts’ stability.
Stability is the second key property that NPMCs still lack compared to Pt/C.
The 2020 target for fuel cells set by the DOE is 5000 h, but in a recent study it was shown that most of the NPMCs lose over 50% of their current in the first 100 h, especially the ones that perform the best at beginning-of-life [39]. The main reasons for activity loss are micropore flooding [87], active site protonation and anion adsorption [71], demetallation [88,89] and oxidation of the carbon material [90]. Since some types of M-Nx sites are known to be situated in micropores [91], creating M-Nx active sites requires nitrogen (which is the main site for protonation) and as said, defects in the carbon plane (which can be oxidised), none of these mechanisms can be avoided completely with the current methods. While these effects are known to be more detrimental in acidic (and thus more corrosive) conditions, they also deteriorate the catalysts in alkaline media [92,93].
4.3.2 Oxygen reduction on metal-free catalysts
The second type of replacement catalysts developed forgoes metals entirely and the focus has been on maximising the activity of nitrogen-doped carbon nano-materials [49,94]. Metal-free catalysts offer a couple of important advantages over metal-based catalysts, as the first mechanism of activity loss (demetalation of the active site) is not present here (thus these catalysts are intrinsically more stable) and also circumvent the use of metals in the synthesis, of which the production of some is actually quite bad for the environment (for example, Mn and Co-based catalysts) [81,95–97].
By introducing electron-rich nitrogen atoms into the carbon support, the π electrons in carbon are conjugated to the lone-pair electrons of nitrogen atom [98]. The effect of nitrogen doping depends highly on the placement of the dopant atoms in the lattice [99–101]. Reportedly there are four different types of nitrogen species in N-doped carbon-based materials, pyridinic-N, pyrrolic-N, quaternary-N (graphitic-N), and pyridinic N+–O−. Although metal-free N-doped carbon catalysts
have become an increasingly researched topic in the last 10 years, the origin of the ORR activity in N-doped carbon materials remains a controversial topic.
There have been a number of debates about this, such as the spatial location (basal or edge sites) or chemical nature (which nitrogen moiety or combination is re-sponsible for the ORR activity). The lone pair of electrons on a pyridinic N atom (a nitrogen atom in a six-member carbon ring bound to two carbon atoms) has long been thought to be the site responsible for most of the ORR activity due to the π-conjugation it forms [102,103]. Studies made with catalysts containing almost exclusively pyridinic N or comparisons with a large number catalysts with different moieties have confirmed the positive effect of pyridinic N on the ORR kinetics [53,102,104–106]. Theoretical calculations as well as studies with real catalysts have proven that graphitic N (a nitrogen atom in a six-member carbon ring bound to three carbon atoms) also contributes to the ORR activity of a N-doped carbon material [107,108]. However, real life catalysts do not comprise of perfect sp2 graphitic sheets, the material is often rather amorphous and defective. Indeed, a high number of defects and edge sites is also a known descriptor for highly active N-doped carbon catalysts for ORR [109]. In addition to this, a large surface area is obviously needed for a high number of active sites on the catalyst surface.
Another important factor for ORR activity in N-doped carbons is the amount of nitrogen both on the surface and in the bulk of the material. Surface concentration of nitrogen of up to 8.4% were shown by Rao et al. to increase the ORR electro-catalytic activity in acidic solutions [110]. In alkaline conditions, similar results have been found [111,112], suggesting that the number of active sites for both alkaline and acidic O2 cleavage rises as the surface nitrogen content increases.
The increase of bulk N content has been shown to have an effect on the valence and conductive bands near the Fermi level, resulting in metallic conductivity [113,114], which in turn increases the catalytic activity as the speed of electron transport rises. Unfortunately, even the best metal-free catalysts have not yet reached the level of electrocatalytic activity seen in M-N-C materials, especially in acidic conditions.